by Dustin Sklavos
So here we are, into the fourth entry in the "How it Works" series, and it feels like we’ve barely scratched the surface. If laptops are so simple, why the heck do they require a massive multipart series like this? Try to keep things in perspective: there are only a few key concepts that you need to understand. Like the motherboard article, for example. You don’t need to know every little thing the motherboard does, just that it connects everything to everything else. Likewise, in the processor article, I’m mostly interested in providing a bit of an easy reference for all the numbers and "useless information" that Intel and AMD are apt to throw at you.
Now that we’ve been going for a couple weeks, though, I do want to address some of the comments on the forums. I really appreciate the support, and it’s gratifying to watch my rep jump up every time a new one of these is posted. That said, on each article there are generally one or two people that want to know why I didn’t cover something, or why I fudged something else.
On the latter part, if there’s a mistake, it’s either an intentional fudging (I usually mark those), or honestly, I’m just human. I may be hanging my cred out to dry here, but I’m a hobbyist/enthusiast first and foremost, and not an engineer. If I do make a mistake, please feel free to correct me in the forums … that’s just one of the things they’re there for. But most of the time, I’m pretty confident I’ve got my facts in order – confident enough that I’ve been able to (and fortunate enough to) continue freelance writing for Notebook Review for as long as I have.
On the former, please note that most of my omissions are intentional ones. I’m keenly aware of how many other hardware enthusiasts frequent Notebook Review, and it’s one of the things that makes it such a nice place to hang out. That said, details that seem elementary to us as enthusiasts can needlessly complicate things, and oftentimes are just plain useless to the people I’m largely trying to reach: consumers like my dad who just don’t know what the heck all this stuff is. I’m not trying to create an intensely detailed series here, I’m just tossing out the basics. So if I omit something, chances are it was intentional. Now, on with the show!
How It Works: Memory
Memory, or RAM (which is just an acronym for "Random Access Memory") is one of the big reasons why "bandwidth" was one of the key terms I defined in Part I. Simply put, memory is a sort of a buffer.
Let me explain: Your processor has its own on-board memory (explained in Part III) which is very fast. Your hard disk, on the other hand, is dog slow. There isn’t a whole lot that can be done about that; even just ignoring the physical mechanics of it, you can think of your processor as being able to store a single newspaper worth of information, and the hard disk as being an archive of every newspaper printed since 1920. It’s a heck of a lot easier to find the obituaries in the single newspaper than the obituaries in one of the hundreds of thousands of newspapers. Simply put, it takes a heck of a lot longer to find it.
So if your processor just had a direct connection to your hard disk, your computer would be dog slow and horribly unresponsive. And this is where RAM comes in.
RAM basically works like a buffer for the processor between any number of the components inside the system. Since they all have varying degrees of speed in which they work, RAM was designed as a high-speed place to stash information and pull it up as needed. So while your hard disk may be peaking at 60MB/sec on a good day, your RAM can be hitting at least 6GB/sec or better. So we’re clear, that’s 100 times faster, and again, these are rough figures and I’m fudging them a bit. 6GB/sec still isn’t as fast as the on-die cache of the processor is, but it’s a hell of a lot faster than waiting for the hard drive.
Before I get into the statistics of memory itself, though, I want to talk a little bit about …
The Memory Interface
AMD and Intel use two different interfaces for accessing RAM. AMD uses what’s called Hypertransport technology, and what that basically means is that the memory is connected directly to the processor instead of going through the northbridge. There are tradeoffs involved (the processor is tied to a specific type of memory and only certain speeds, and the memory needs to be in close physical proximity to the processor), but the flipside is a massive boost in memory bandwidth.
Intel, on the other hand, uses a front-side bus, which I discussed in previous articles. While the front-side bus technology has evolved quite well, it still can’t quite match the kind of raw bandwidth Hypertransport is able to produce; as a result, Intel’s next chip design ditches the front-side bus to use an Intel-developed analogue to Hypertransport, a technology they call Quickpath. The same principles as Hypertransport still apply, though.
Of course, something else needs to be kept in mind: a boost in memory bandwidth doesn’t mean anything if the processor itself isn’t fast enough to take advantage of it. This is going to be a reoccurring theme at every point in this article: the value of memory will often be dependent on the performance of the processor itself. So just because AMD’s chips enjoy a substantially fatter pipeline to their memory doesn’t automatically make them faster; the processor itself has to be able to actually utilize all that bandwidth efficiently.
Okay, so basically there are three types of memory in circulation right now: DDR, DDR2, and DDR3. The "DDR" stands for "Double Data Rate," but don’t concern yourself with that; all that really means is that the effective speed of the memory is twice the actual clock speed it runs at. Because, yes, memory runs at a clock speed, just like your processor does, and that speed is actually tied to your processor’s speed. It’s a little complicated, but I’ll explain.
So what do you need to know about all these memory types? Simple: they aren’t interchangeable, they aren’t backwards or forwards compatible, and DDR2 is the most pervasive one as of the writing of this article. Consider them sequels to each other, revisions and redesigns that improve on their predecessors.
Now remember what I said about Hypertransport and how it ties the processor to a specific type of memory? This is what I’m talking about. Intel can produce a northbridge that communicates with both DDR2 and DDR3 (though never simultaneously). AMD, however, is tied to DDR2. In order for an AMD chip to support DDR3, that chip has to be physically designed to support it. Likewise, when Intel makes the switch to Quickpath, that switch will prevent them from being as flexible with memory types as they are today.
I’m gonna let you in on a big secret: memory speed is almost totally irrelevant to your system’s performance and has virtually no tangible effect on its day to day use. The exception would be integrated graphics parts which just appreciate all the memory bandwidth they can get.
The speed of memory is defined in two ways: its clock speed, and its theoretical bandwidth. So let’s say you have DDR2 memory that runs at 667 MHz, the common speed for modern notebook RAM. That memory would be notated as DDR2-667. However, it may also be notated by its theoretical bandwidth, in which case it would be PC2-5400 or PC2-5300. This is where it gets a little bit silly: some manufacturers prefer one over the other or may play up this number. Mercifully, however, the vast majority of people just prefer to notate memory the first way, as its speed.
At this point I’m going to get into a little bit of minutiae that may or may not be of value to you: the listed speed of the memory is its specification speed. This means that DDR2-800 is specced to run UP TO DDR2-800, but can also run happily at lower speeds.
The really crazy details follow, and you’re welcome to skip these. If you’re not interested in getting a brain cramp, just cut to the part that says "END BRAIN CRAMP" and move forward.
There’s basically a relationship between memory speed, front-side bus speed, and processor speed. It’s okay, it’s a lot simpler than it sounds. First of all, for DDR2, divide the noted clock speed by four to get its actual clock speed. So DDR2-667, for example, is actually running at 166 MHz. On an Intel machine, the front-side bus speed is defined the same way. And then the speed of the processor is defined by the front-side bus’s true clock speed multiplied by what’s called a – wait for it – "multiplier." And the speed the system runs the memory at is defined by a ratio against the front side bus speed.
That’s a whole lot of gobbledegook, so let’s give you an example.
You have a 2.4GHz processor with an 800MHz front-side bus. What this means is:
The front-side bus’s true clock speed is 200MHz, but because Intel "quad pumps" the front-side bus, the speed is 800MHz effectively, and that’s how it’s referred to.
The processor’s clock speed, 2.4GHz, is achieved by multiplying that front-side bus speed (200MHz) by the processor’s multiplier (in this case, 12). Presto, 2.4GHz.
Now let’s say you’re using DDR2-800. The northbridge sets the memory ratio at 1:1, so the memory is basically running at the same speed as the front-side bus.
Of course, it doesn’t always work that way, so memory may run at a different ratio, like 6:4, or 12:8 as is the case on my desktop machine.
** END BRAIN CRAMP **
That mess above is provided you’re running an Intel machine; AMD’s HyperTransport will make memory run at odd frequencies. Don’t worry about it.
The important thing to remember is that on AMD machines, the memory speed makes a small difference in performance, and on Intel machines, it makes almost no difference.
A Brief Word About Timings
Someone will someday talk to you about memory timings. This is esoterica you don’t need to worry about or give a crap about. Memory timings are for enthusiasts with too much free time and people who think they’re actually achieving something.
I’ve tested standard Corsair RAM and I’ve tested high end OCZ and Crucial RAM. You can tease out differences based on memory speed and timings in synthetic benchmarks, but in practical use it’s virtually impossible to tell the difference. Enthusiast RAM exists to separate suckers from their money.
Dual and Single Channel
This is another brief word and it’s a simple one. Your notebook is typically designed to handle memory in a configuration called "dual channel." Without getting into too much detail, it’s basically a means of effectively doubling memory bandwidth by running a matched pair of memory sticks.
In practice, the difference between single and dual channel is largely negligible on Intel platforms (just like the speed), though it’s much more relevant on AMD systems.
Intel systems are also designed to operate in a sort of "hybrid" fashion in such a way that two unmatched sticks of RAM can run in a weird dual channel mode, so if you have a 1GB stick and a 2GB stick that are running at the same speed, you can get most of the benefit of dual channel operation.
Ultimately, this is performance that you don’t need to pay extra for, and it’s too easy to pick up for you to leave it on the table.
This is, of course, "the big one." First of all, sticks of memory are widely available in the following capacities: 256MB, 512MB, 1GB (1024MB), and 2GB (2048MB).
This is the aspect of memory that really affects performance, because this is how much your computer can actually access fairly quickly before it winds up hitting the hard disk. When you run a program, that program is loaded into RAM. The more RAM you have, the more that can be loaded into RAM, and thus the faster the laptop can run. If you start to run out of RAM, your system will hit what’s called the "page file:" basically an optimized portion of the hard disk used to stash stuff that won’t immediately fit in memory but may need to be quickly swapped in later. You don’t want to hit the page file.
However, shedloads of RAM aren’t automatically going to make your computer run faster. In an Atom-based machine, for example, more than 1GB of RAM is a waste; the Atom just isn’t fast enough to really make use of all of it. A faster processor – particularly the quad cores on the horizon – will more readily use a greater amount of memory.
The amount of memory you need also depends on the operating system you’re running. If you’re using 32-bit Windows Vista, 2GB is really as low as you want to go. In 64-bit Vista, 4GB is ideal. And if you’re using Windows XP, 1GB is the minimum, and again, 2GB is preferable. As for Mac OS X or Linux? Mac OS X will use as much RAM as you can throw at it, and tends to be a bit of a hog, so I’d go for at least 2GB. And if you’re running some flavor of Linux, mercifully Linux is pretty efficient, so you’ll probably be fine with even just 1GB.
And, per the "dual channel" bit before, if you’re going to run 2GB, ideally you’ll want to do it in a pair of 1GB sticks as opposed to a single 2GB. While the single gives you an upgrade path later on, it leaves a little performance on the table. If you’re thinking about upgrading later, just do it now and get it out of the way. RAM is cheap and easy to add.
But since we’re talking about upgrading …
First of all, if you buy your laptop custom built, don’t pay extra for RAM. Manufacturers – especially Apple – will gouge you like crazy on it. Just get the bare minimum; RAM can be bought very cheaply online at sites like NewEgg. I’d advise against buying in the store either; Best Buy in particular gouges even worse than the manufacturers do, charging as much as three times more than you’d pay buying it online.
Now, I have yet to see a laptop that comes with more than two memory slots, so while your desktop may happily accommodate 8GB (as mine does), your laptop’s going to top out at 4GB. MOST (not all) laptops have a user-accessible panel on the bottom of the machine that lets you access the memory slots, and a careful inspection of the slots inside will show how the memory is held in. In laptops in particular, you generally insert the RAM diagonally as opposed to vertically or horizontally: you tilt it into the memory slot then push it down until the retaining clips snap into place.
If you’re worried about this voiding your warranty, you can (and should) of course check what your warranty covers or contact customer service, but the vast majority of modern notebooks will allow you to upgrade the memory or hard disk without voiding the warranty.
Recommendations and Conclusion
This was a much more long-winded installment than my previous entries. As far as memory goes, as I said before, buy only online or at a place which offers prices comparable to buying online. For example, as of this writing the only 4GB DDR2 laptop memory kit I could find on Best Buy’s site is going for $99 – on sale from $319. A visit to NewEgg has the same kit for just $70, and that’s not even for a particularly great brand.
If you’re buying memory, the most trouble free brand I can recommend is Corsair (though Crucial, Kingston, and Patriot can also be solid buys). I happily run Corsair in my desktop, and it’s basically become the stand-by for reliability in my computers.
Now this has been a lot to digest and unfortunately, I can’t really break it down into easy to digest points for you because there are so many little conditions here and there. Memory is a big deal, and the only points I can really give you are:
- More is generally better, so you might as well max out your laptop if you’re planning on keeping it for a while. Even if you’re only running a 32-bit operating system, 4GB is still my recommendation.
- Memory is one of those things that, like additional CPU cores, doesn’t make your system FASTER so much as SMOOTHER.
In my next "How it Works" article, I’ll finally tackle the one you’ve all been waiting for (or at least a lot of you have been waiting for): graphics hardware.
Thanks for reading!